Presentation
Machine Learning Noninvasive Intracranial Pressure in the Future of Neurocritical Care
DescriptionIn recent years, the potential of computerized machine learning (ML) processes has been widely explored in clinical research, with promising results. Regarding acute brain injured (ABI) patients, the establishment of extensive databases from the collection of diverse biometric markers using various neuromonitoring techniques has enabled the advancement of ML models, transforming their diagnostic approach.
The present session is dedicated to present and discuss ML applications in neuro-ICU, in the point-of-view of three clinicians involved with model development, research and translation to clinical practice.
Lecture 1- Noninvasive estimation of intracranial pressure (ICP); A large dataset of ICP recording was collected from 146 ABI patients using both invasive (external ventricular drains) and noninvasive systems concomitantly. The noninvasive system is based on skull micro-expansions, registering surrogate ICP pulse morphology. The ML model developed allowed estimating ICP values from noninvasive pulse slopes accurately, providing a pioneer means of ICP monitoring exempt of additional risks for patients.
Lecture 2- Early detection of elevated risk for ischemia after spontaneous subarachnoid hemorrhage (SAH); A ML model was created to predict delayed cerebral ischemia (DCI) in 399 spontaneous SAH patients by means of a comprehensive eletronic medical record including radiological, laboratorial and clinical data. The model was succesful to predict DCI as well as discharge and 3-month outcomes, with potential to improve SAH management.
Lecture 3- A ML approach combining information from transcranial Doppler, arterial blood pressure and electrocardiogram was built to indicate ideal cerebral perfusion pressure in ABI patients, leveling cerebral hemodynamic autoregulation failure in this population.
The present session is dedicated to present and discuss ML applications in neuro-ICU, in the point-of-view of three clinicians involved with model development, research and translation to clinical practice.
Lecture 1- Noninvasive estimation of intracranial pressure (ICP); A large dataset of ICP recording was collected from 146 ABI patients using both invasive (external ventricular drains) and noninvasive systems concomitantly. The noninvasive system is based on skull micro-expansions, registering surrogate ICP pulse morphology. The ML model developed allowed estimating ICP values from noninvasive pulse slopes accurately, providing a pioneer means of ICP monitoring exempt of additional risks for patients.
Lecture 2- Early detection of elevated risk for ischemia after spontaneous subarachnoid hemorrhage (SAH); A ML model was created to predict delayed cerebral ischemia (DCI) in 399 spontaneous SAH patients by means of a comprehensive eletronic medical record including radiological, laboratorial and clinical data. The model was succesful to predict DCI as well as discharge and 3-month outcomes, with potential to improve SAH management.
Lecture 3- A ML approach combining information from transcranial Doppler, arterial blood pressure and electrocardiogram was built to indicate ideal cerebral perfusion pressure in ABI patients, leveling cerebral hemodynamic autoregulation failure in this population.
Speaker
Event Type
Breakout Session
TimeTuesday, October 15th1:30pm - 1:50pm PDT
LocationHarbor Ballrooms D-I
Science of Neurocritical Care
General Critical Care
Global Neurocritical Care
Informatics
Multimodal Neuromonitoring (invasive/non-invasive)
Subarachnoid Hemorrhage
Traumatic Brain Injury
Intermediate